Neil Schemenauer

> These tests (and the PR) has N = 1'000'000. The downside is that running the benchmark (with Python 3.14) takes almost 10 minutes. Taking 10 minutes would be too...

Small suggestion: it would be simpler to use `io.BytesIO()` rather than using real files in a temporary folder. I don't think that affects the usefulness of the benchmark, since we...

A benchmark based on [pickletools](https://github.com/python/cpython/issues/140175) might be good. It seems to show this regression.

Some additional info: 3.13 memory use stable, 3.14.1 grows, 3.14.2 grows, 3.14.2t grows. I initially suspected this would be fixed by https://github.com/python/cpython/pull/142001 but since it leaks for both the 3.14.1...

I tried to narrow down the issue a bit better. It seems SSL is not required. I'm running the script below. It loads from a local web server (port 8001...

I was wrong about SSL not being involved. Here is a memray flame graph. Nearly all the memory is allocated from `load_verify_locations()`, which is implemented in the `_ssl.c` module, it...

> One example of the cycle: > > Response -> BoundAsyncStream -> Response -> ... Good detective work! If I break that cycle, the high memory usage goes away. E.g....

I don't know if this is an accurate minimization of the test case but it does use more memory in 3.14 vs 3.13 (and more yet in 3.14.2). It's just...

> I'm interested on this issue. > Could I have a try? 😊 Sure. Note that there was some discussion about the best way to fix this on the free-threading...